Copied tests and some definitions from base

[ararslan]

Let's see what happens to the coverage and CI...

[ararslan]

...the answer appears to be "not good things."

In this repo we're doing some of the bessel return values with oftype(x, ...), which is omitted in Base. Should we just copy what Base is doing or is there a good reason to be doing it this way? cc @tkelman

[tkelman]

I'm not sure! What is the type difference if you leave out the oftype? Can you see anything in the history of the base files where maybe it was changed at some point that this repo missed?

[ararslan]

I can check the return type differences a little later but in the meantime I was able to find the commit that removed the oftypes: JuliaLang/julia@8077d11

@simonbyrne As the one who introduced the change, can you provide some insight into this?

[simonbyrne]

As the commit comment mentions, it was to fix JuliaLang/julia#11512

[ararslan]

Right, that fixed the reported issue with BigFloats, I was just wondering if there was any other justification for that particular solution since removing oftype seems like it would make the return type less predictable in general. (Maybe that's totally wrong, I don't know.)

[ararslan]

In particular, it looks like one of the test failures was because oftype was converting the return type, which would otherwise have been a float, to complex to match the argument's type. No idea whether that's desirable.

[ararslan]

First of all, sorry for the month and a half delay in getting this done.

I think I had misinterpreted the errors before; I don't think Simon's removal of oftype was actually causing the type difference I was seeing. It appears it was a thing with the 0.4 version of isapprox that Simon has since fixed in base. Sorry for the confusion, and thanks to Simon for fixing isapprox!

I worked around the isapprox difference in here and got all AMOS-based code and tests copied over from base.